[ath9k-devel] [PATCH] ath9k: Prevent radar detection and spectral scan to be used concurrently

On 11/21/2016 04:10 PM, Michal Kazior wrote:
> On 21 November 2016 at 15:41, Zefir Kurtisi <[email protected]> wrote:
>> On 11/21/2016 03:04 PM, Benjamin Berg wrote:
>>> In the case that a spectral scan is enabled the PHY errors sent by the
>>> hardware as part of the scanning might trigger the radar detection and
>>> channels might be marked as 'unusable' incorrectly. This patch fixes
>>> the issue by preventing the spectral scan to be enabled if DFS is used
>>> and only analysing the PHY errors for DFS if radar detection is enabled.
>>>
>>> [...]
>>
>> From the relevant source code portion in channel.c:ath_set_channel()
>>
>> 80         /* Enable radar pulse detection if on a DFS channel. Spectral
>> 81          * scanning and radar detection can not be used concurrently.
>> 82          */
>> 83         if (hw->conf.radar_enabled) {
>> 84                 u32 rxfilter;
>> 85
>> 86                 rxfilter = ath9k_hw_getrxfilter(ah);
>> 87                 rxfilter |= ATH9K_RX_FILTER_PHYRADAR |
>> 88                                 ATH9K_RX_FILTER_PHYERR;
>> 89                 ath9k_hw_setrxfilter(ah, rxfilter);
>> 90                 ath_dbg(common, DFS, "DFS enabled at freq %d\n",
>> 91                         chan->center_freq);
>> 92         } else {
>> 93                 /* perform spectral scan if requested. */
>> 94                 if (test_bit(ATH_OP_SCANNING, &common->op_flags) &&
>> 95                         sc->spec_priv.spectral_mode == SPECTRAL_CHANSCAN)
>> 96                         ath9k_cmn_spectral_scan_trigger(common, &sc->spec_priv);
>> 97         }
>>
>> it seems that spectral can't ever be activated while operating on a DFS channel.
>>
>> If you need to catch the opposite case, i.e. prevent feeding pseudo-radar pulses
>> into the pattern detector, you just need to ensure that they depend on
>> hw->conf.radar_enabled. A patch like the attached one should be enough.
> 
> Good point. I guess set_channel could be oversimplified as well. I
> mean, it makes sense to consider radar and spectral mutually exclusive
> if they use the same phyerr code. However some chips actually seem (as
> per the comment I mentioned) to distinguish the two so I don't know if
> the "mutually exclusive" is true for all chips per se. Just thinking
> out loud.
> 
You're right, blindly feeding PHYERR frames to spectral when spectral is not
enabled is as wrong as feeding them to the dfs-detector when not on a radar
channel. Therefore, the patch I proposed should be extended to make calling
ath_cmn_process_fft() depending on
a) not operating on radar channel, and
b) spectral scan is enabled

Updated patch attached as proposal.

> I also wonder if calling ieee80211_radar_detect() should have any
> effect if there are no radar operated interfaces in the first place?
> 
True, calling ieee80211_radar_detect() for a non-DFS channel is (or should be)
ignored by mac. But feeding the dfs-detector with invalid pulses causes quite some
computational overhead - dropping them if no radar detection is required is a good
thing to do.


Cheers,
Zefir




-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-ath9k-feed-DFS-detector-only-if-operating-on-radar-c.patch
Type: text/x-patch
Size: 0 bytes
Desc: not available
Url : http://lists.ath9k.org/pipermail/ath9k-devel/attachments/20161121/ab36ab4e/attachment-0001.bin